Chuck reported [1] an IO hang problem on NFS exports that reside on SATA
devices and bisected to commit 615939a2ae ("blk-mq: defer to the normal
submission path for post-flush requests").
We analysed the IO hang problem, found there are two postflush requests
waiting for each other.
The first postflush request completed the REQ_FSEQ_DATA sequence, so go to
the REQ_FSEQ_POSTFLUSH sequence and added in the flush pending list, but
failed to blk_kick_flush() because of the second postflush request which
is inflight waiting in scheduler queue.
The second postflush waiting in scheduler queue can't be dispatched because
the first postflush hasn't released scheduler resource even though it has
completed by itself.
Fix it by releasing scheduler resource when the first postflush request
completed, so the second postflush can be dispatched and completed, then
make blk_kick_flush() succeed.
While at it, remove the check for e->ops.finish_request, as all
schedulers set that. Reaffirm this requirement by adding a WARN_ON_ONCE()
at scheduler registration time, just like we do for insert_requests and
dispatch_request.
[1] https://lore.kernel.org/all/7A57C7AE-A51A-4254-888B-FE15CA21F9E9@oracle.com/
Link: https://lore.kernel.org/linux-block/20230819031206.2744005-1-chengming.zhou@linux.dev/
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202308172100.8ce4b853-oliver.sang@intel.com
Fixes: 615939a2ae ("blk-mq: defer to the normal submission path for post-flush requests")
Reported-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Tested-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/20230813152325.3017343-1-chengming.zhou@linux.dev
[axboe: folded in incremental fix and added tags]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit ee6d3dd4ed ("driver core: make kobj_type constant.")
the driver core allows the usage of const struct kobj_type.
Take advantage of this to constify the structure definitions to prevent
modification at runtime.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Link: https://lore.kernel.org/r/20230208-kobj_type-block-v1-1-0b3eafd7d983@weissschuh.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The kobject embedded into the request_queue is used for the queue
directory in sysfs, but that is a child of the gendisks directory and is
intimately tied to it. Move this kobject to the gendisk and use a
refcount_t in the request_queue for the actual request_queue refcounting
that is completely unrelated to the device model.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221114042637.1009333-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove the description of @required_features in elevator_match()
to clear the below warning:
block/elevator.c:103: warning: Excess function parameter 'required_features' description in 'elevator_match'
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=2734
Fixes: ffb86425ee ("block: don't check for required features in elevator_match")
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Link: https://lore.kernel.org/r/20221107062255.2685-1-yang.lee@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split an elevator_disable helper from elevator_switch for the case where
we want to switch to no scheduler at all. This includes removing the
pointless elevator_switch_mq helper and removing the switch to no
schedule logic from blk_mq_init_sched.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Checking for the required features in the callers simplifies the code
quite a bit, so do that.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-7-hch@lst.de
[axboe: adjust for dropping patch 1, use __elevator_find()]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use eq for the elevator_queue as done elsewhere. This frees e to be used
for the loop iterator instead of the odd __ prefix. In addition rename
elv to cur to make it more clear it is the currently selected elevator.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If the tag_set has BLK_MQ_F_NO_SCHED flag set we will never show any
scheduler, so exit early.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Do the request_module and repeated lookup in the only caller that cares,
pick a saner name that explains where are actually doing a lookup and
use a sane calling conventions that passes the queue first.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The current reference management logic of io scheduler modules contains
refcnt problems. For example, blk_mq_init_sched may fail before or after
the calling of e->ops.init_sched. If it fails before the calling, it does
nothing to the reference to the io scheduler module. But if it fails after
the calling, it releases the reference by calling kobject_put(&eq->kobj).
As the callers of blk_mq_init_sched can't know exactly where the failure
happens, they can't handle the reference to the io scheduler module
properly: releasing the reference on failure results in double-release if
blk_mq_init_sched has released it, and not releasing the reference results
in ghost reference if blk_mq_init_sched did not release it either.
The same problem also exists in io schedulers' init_sched implementations.
We can address the problem by adding releasing statements to the error
handling procedures of blk_mq_init_sched and init_sched implementations.
But that is counterintuitive and requires modifications to existing io
schedulers.
Instead, We make elevator_alloc get the io scheduler module references
that will be released by elevator_release. And then, we match each
elevator_get with an elevator_put. Therefore, each reference to an io
scheduler module explicitly has its own getter and releaser, and we no
longer need to worry about the refcnt problems.
The bugs and the patch can be validated with tools here:
https://github.com/nickyc975/linux_elv_refcnt_bug.git
[hch: split out a few bits into separate patches, use a non-try
module_get in elevator_alloc]
Signed-off-by: Jinlong Chen <nickyc975@zju.edu.cn>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221020064819.1469928-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
No need to find the actual elevator_type struct for this comparism,
the name is all that is needed.
Signed-off-by: Jinlong Chen <nickyc975@zju.edu.cn>
[hch: split from a larger patch]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221020064819.1469928-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The stripped name should also be used for the none check. To do so
strip it in the caller and pass in the sanitized name. Drop the pointless
__ prefix in the function name while we're at it.
Based on a patch from Jinlong Chen <nickyc975@zju.edu.cn>.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221020064819.1469928-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make sure we have helpers for all relevant module refcount operations on
the elevator_type in elevator.h, and use them. Move the call to the get
helper in blk_mq_elv_switch_none a bit so that it is obvious with a less
verbose comment.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221020064819.1469928-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There are only one flag to indicate that elevator is registered currently,
prepare to add a flag to disable wbt if default elevator is bfq.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221019121518.3865235-6-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
"elevator_queue *e" is already declared and initialized in the beginning
of elv_unregister_queue().
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221019121518.3865235-2-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The hctx's run_work may be racing with the elevator switch when
reinitializing hardware queues. The queue is merely frozen in this
context, but that only prevents requests from allocating and doesn't
stop the hctx work from running. The work may get an elevator pointer
that's being torn down, and can result in use-after-free errors and
kernel panics (example below). Use the quiesced elevator switch instead,
and make the previous one static since it is now only used locally.
nvme nvme0: resetting controller
nvme nvme0: 32/0/0 default/read/poll queues
BUG: kernel NULL pointer dereference, address: 0000000000000008
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 80000020c8861067 P4D 80000020c8861067 PUD 250f8c8067 PMD 0
Oops: 0000 [#1] SMP PTI
Workqueue: kblockd blk_mq_run_work_fn
RIP: 0010:kyber_has_work+0x29/0x70
...
Call Trace:
__blk_mq_do_dispatch_sched+0x83/0x2b0
__blk_mq_sched_dispatch_requests+0x12e/0x170
blk_mq_sched_dispatch_requests+0x30/0x60
__blk_mq_run_hw_queue+0x2b/0x50
process_one_work+0x1ef/0x380
worker_thread+0x2d/0x3e0
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220927155652.3260724-1-kbusch@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move the calls to ioc_clear_queue and blk_mq_sched_free_rqs into
elevator_exit. Except for one call where we know we can't have io_cq
structures yet these always go together, and that extra call in an
error path is harmless.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220308055200.735835-14-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make elv_unregister_queue() a no-op if q->elevator is NULL or is not
registered.
This simplifies the existing callers, as well as the future caller in
the error path of blk_register_queue().
Also don't bother checking whether q is NULL, since it never is.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220124215938.2769-2-ebiggers@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that we disable wbt by set WBT_STATE_OFF_DEFAULT in
wbt_disable_default() when switch elevator to bfq. And when
we remove scsi device, wbt will be enabled by wbt_enable_default.
If it become false positive between wbt_wait() and wbt_track()
when submit write request.
The following is the scenario that triggered the problem.
T1 T2 T3
elevator_switch_mq
bfq_init_queue
wbt_disable_default <= Set
rwb->enable_state (OFF)
Submit_bio
blk_mq_make_request
rq_qos_throttle
<= rwb->enable_state (OFF)
scsi_remove_device
sd_remove
del_gendisk
blk_unregister_queue
elv_unregister_queue
wbt_enable_default
<= Set rwb->enable_state (ON)
q_qos_track
<= rwb->enable_state (ON)
^^^^^^ this request will mark WBT_TRACKED without inflight add and will
lead to drop rqw->inflight to -1 in wbt_done() which will trigger IO hung.
Fix this by move wbt_enable_default() from elv_unregister to
bfq_exit_queue(). Only re-enable wbt when bfq exit.
Fixes: 76a8040817 ("blk-wbt: make sure throttle is enabled properly")
Remove oneline stale comment, and kill one oneshot local variable.
Signed-off-by: Ming Lei <ming.lei@rehdat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/linux-block/20211214133103.551813-1-qiulaibin@huawei.com/
Signed-off-by: Laibin Qiu <qiulaibin@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Partition include/linux/blk-cgroup.h into two parts: one is public part,
the other is block layer private part.
Suggested by Christoph Hellwig.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220211101149.2368042-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Open code elevator_exit in it's only caller, and rename __elevator_exit to
elevator_exit.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20211123185312.1432157-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
elevator_init_mq() is only called before adding disk, when there isn't
any FS I/O, only passthrough requests can be queued, so freezing queue
plus canceling dispatch work is enough to drain any dispatch activities,
then we can avoid synchronize_srcu() in blk_mq_quiesce_queue().
Long boot latency issue can be fixed in case of lots of disks added
during booting.
Fixes: 737eb78e82 ("block: Delay default elevator initialization")
Reported-by: yangerkun <yangerkun@huawei.com>
Cc: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20211117115502.1600950-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Except for the features passed to blk_queue_required_elevator_features,
elevator.h is only needed internally to the block layer. Move the
ELEVATOR_F_* definitions to blkdev.h, and the move elevator.h to
block/, dropping all the spurious includes outside of that.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20210920123328.1399408-13-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When merging one bio to request, if they are discard IO and the queue
supports multi-range discard, we need to return ELEVATOR_DISCARD_MERGE
because both block core and related drivers(nvme, virtio-blk) doesn't
handle mixed discard io merge(traditional IO merge together with
discard merge) well.
Fix the issue by returning ELEVATOR_DISCARD_MERGE in this situation,
so both blk-mq and drivers just need to handle multi-range discard.
Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Fixes: 2705dfb209 ("block: fix discard request merge")
Link: https://lore.kernel.org/r/20210729034226.1591070-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that device mapper has been changed to register the disk once
it is fully ready all this code is unused.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Link: https://lore.kernel.org/r/20210804094147.459763-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
elevator_get_default() uses the following algorithm to select an I/O
scheduler from inside add_disk():
- In case of a single hardware queue or if sharing hardware queues across
multiple request queues (BLK_MQ_F_TAG_HCTX_SHARED), use mq-deadline.
- Otherwise, use 'none'.
This is a good choice for most but not for all block drivers. Make it
possible to override the selection of mq-deadline with a new flag,
namely BLK_MQ_F_NO_SCHED_BY_DEFAULT.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Martijn Coenen <maco@android.com>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20210805174200.3250718-2-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Lockdep complains about lock inversion between ioc->lock and bfqd->lock:
bfqd -> ioc:
put_io_context+0x33/0x90 -> ioc->lock grabbed
blk_mq_free_request+0x51/0x140
blk_put_request+0xe/0x10
blk_attempt_req_merge+0x1d/0x30
elv_attempt_insert_merge+0x56/0xa0
blk_mq_sched_try_insert_merge+0x4b/0x60
bfq_insert_requests+0x9e/0x18c0 -> bfqd->lock grabbed
blk_mq_sched_insert_requests+0xd6/0x2b0
blk_mq_flush_plug_list+0x154/0x280
blk_finish_plug+0x40/0x60
ext4_writepages+0x696/0x1320
do_writepages+0x1c/0x80
__filemap_fdatawrite_range+0xd7/0x120
sync_file_range+0xac/0xf0
ioc->bfqd:
bfq_exit_icq+0xa3/0xe0 -> bfqd->lock grabbed
put_io_context_active+0x78/0xb0 -> ioc->lock grabbed
exit_io_context+0x48/0x50
do_exit+0x7e9/0xdd0
do_group_exit+0x54/0xc0
To avoid this inversion we change blk_mq_sched_try_insert_merge() to not
free the merged request but rather leave that upto the caller similarly
to blk_mq_sched_try_merge(). And in bfq_insert_requests() we make sure
to free all the merged requests after dropping bfqd->lock.
Fixes: aee69d78de ("block, bfq: introduce the BFQ-v0 I/O scheduler as an extra scheduler")
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210623093634.27879-3-jack@suse.cz
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The insert_requests and dispatch_request elevator operations are
mandatory for the correct execution of an elevator, and all implemented
elevators (bfq, kyber and mq-deadline) implement them. As a result,
there is no need to check for these operations before calling them when
a queue has an elevator set. This simplifies the code in
__blk_mq_sched_dispatch_requests() and blk_mq_sched_insert_request().
To avoid out-of-tree elevators to crash the kernel in case of bad
implementation, add a check in elv_register() to verify that these
operations are implemented.
A small, probably not significant, IOPS improvement of 0.1% is observed
with this patch applied (4.117 MIOPS to 4.123 MIOPS, average of 20 fio
runs doing 4K random direct reads with psync and 32 jobs).
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20210618015922.713999-1-damien.lemoal@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Don't return the passed in request_queue but a normal error code, and
drop the elevator_init argument in favor of just calling elevator_init_mq
directly from dm-rq.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210602065345.355274-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Yanhui found that write performance is degraded a lot after applying
hctx shared tagset on one test machine with megaraid_sas. And turns out
it is caused by none scheduler which becomes default elevator caused by
hctx shared tagset patchset.
Given more scsi HBAs will apply hctx shared tagset, and the similar
performance exists for them too.
So keep previous behavior by still using default mq-deadline for queues
which apply hctx shared tagset, just like before.
Fixes: 32bc15afed ("blk-mq: Facilitate a shared sbitmap per tagset")
Reported-by: Yanhui Ma <yama@redhat.com>
Cc: John Garry <john.garry@huawei.com>
Cc: Hannes Reinecke <hare@suse.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: John Garry <john.garry@huawei.com>
Link: https://lore.kernel.org/r/20210406031933.767228-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After commit b89f625e28 ("block: don't release queue's sysfs
lock during switching elevator"), whole elevator register and
unregister function are covered by sysfs_lock. So, remove wrong
comment and add lockdep assert.
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have defined common interface blk_queue_registered() to
test QUEUE_FLAG_REGISTERED. Just use it.
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
elv_support_iosched() will check queue_is_mq() for us. So, remove
the redundant check to clean code.
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We will register debugfs for scheduler no matter whether it have
defined callback funciton .exit_sched. So, blk_mq_exit_sched()
is always needed to unregister debugfs. Also, q->elevator should
be set as NULL after exiting scheduler.
For now, since all register scheduler have defined .exit_sched,
it will not cause any actual problem. But It will be more reasonable
to do this change.
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Drop the repeated word "the".
Fix typos of "features" and "specified".
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull on for-linus to resolve what otherwise would have been a conflict
with the cgroups rstat patchset from Tejun.
* for-linus: (942 commits)
blkcg: make blkcg_print_stat() print stats only for online blkgs
nvme: change nvme_passthru_cmd64 to explicitly mark rsvd
nvme-multipath: fix crash in nvme_mpath_clear_ctrl_paths
nvme-rdma: fix a segmentation fault during module unload
iocost: don't nest spin_lock_irq in ioc_weight_write()
io_uring: ensure we clear io_kiocb->result before each issue
um-ubd: Entrust re-queue to the upper layers
nvme-multipath: remove unused groups_only mode in ana log
nvme-multipath: fix possible io hang after ctrl reconnect
io_uring: don't touch ctx in setup after ring fd install
io_uring: Fix leaked shadow_req
Linux 5.4-rc5
riscv: cleanup do_trap_break
nbd: verify socket is supported during setup
ata: libahci_platform: Fix regulator_get_optional() misuse
nbd: handle racing with error'ed out commands
nbd: protect cmd->status with cmd->lock
io_uring: fix bad inflight accounting for SETUP_IOPOLL|SETUP_SQTHREAD
io_uring: used cached copies of sq->dropped and cq->overflow
ARM: dts: stm32: relax qspi pins slew-rate for stm32mp157
...
